Psychology 280: Psychological Statistics Fall 2005 Borton & Pierce p. 1

SPSS Instructions for Research Project #2 (Q-aire Construction)

Paper Due Date: October 5, 2005 Part I: Assessing Internal Reliability You will be using SPSS to calculate a measure of internal reliability called “Cronbach’s α.” Cronbach’s α is a reliability coefficient based on the average covariance among items in a scale. We assume that items on a scale are positively correlated with each other because they’re all tapping into the same construct; that is, they’re all measuring a common entity (some items may need to be reverse scored so that they are all positively correlated). The average correlation of an item with ALL OTHER ITEMS IN THE SCALE tells us about the extent of the common entity. Because α can be interpreted as a correlation coefficient, it ranges in value from 0 to 1. (Negative α values can occur when items aren’t positively correlated among themselves and the reliability model is violated.) Before you can analyze your scale’s reliability, however, you must do the following: 1. Check for errors! One quick way to check for major mistakes is to run a frequency analysis on each

variable. (See other SPSS instruction handout for how to do this.) A frequency will tell you, for each question, how many people responded “1,” how many people responded “2,” and so on. If you find that one person responded “9,” you’ll know you’ve made an error if 9 wasn’t one of the choices. By examining your output, you can also detect if any of your items had ceiling or floor effects.

2. REMEMBER TO RECODE THE QUESTIONS THAT ARE NEGATIVELY KEYED! To do so,

follow these instructions for using the “recode” command:

As we discussed in class, questionnaires are often constructed such that approximately half of the questions are framed positively and the other half are framed negatively. For example, in Rosenberg’s Self-Esteem scale (one of the most widely used measures of self-esteem), items 1, 2, 4, 6, and 7 are written so that high numbers correspond to high self-esteem (e.g., Item 7: “On the whole, I am satisfied with myself”) whereas items 3, 5, 8, 9, and 10 are written so that high numbers correspond to low self-esteem (e.g., Item 3: “All in all, I am inclined to feel that I am a failure”). If we wanted to compute an average self-esteem score for each participant based on his/her responses to all 10 items on the scale, we would need to reverse-score half the items so that they were all coded in the same direction. To do this, we would go up to the “Transform” menu and scroll down to “Recode”. You get two options: “Into same variables” and “Into different variables.” Choose “Into different variables.” You should get the window you see below:

Figure 1

Psychology 280: Psychological Statistics Fall 2005 Borton & Pierce p. 2

The names of all your variables will appear in the list on the left part of the window. Scroll down until you find the variables you’re looking for. In this case, we want to recode se3, se5, se8, se9, and se10. Simply click on the variable you want, then click on the arrow (which will be pointing the opposite of the way it’s pointing in the figure). Do this for all the variables you want to recode the same way. Now you need to tell the computer what you want the new (recoded) variables to be called. To make life easier, I usually just add “r” to the original variable name, to remind myself that the variables have been recoded. So, for example, “se3” becomes “se3r.” Type this new name in the “output variable” box on the right of the window and click on the “Change” button. Do this for all the variables you need to recode. Now you’re ready to tell the computer how to recode the variables. Since the Rosenberg Self-Esteem Scale uses a 4-point scale, you need to recode in the following way: 1=4, 2=3, 3=2, and 4=1. That is, in order for a response of “1” on question 3 to indicate high self-esteem, it needs to be recoded as a “4” (you’re just reversing the scale). To do this, first click on the “Old and New Values” box. You’ll get the screen you see below:

Figure 2 Note how I’ve already recoded 1 into 4; you can see it on the right side of the window in the “OldNew” box. To change 2 into 3, simply type 2 in the “Old Value” box on the left and type 3 in the “New Value” box on the right. Then click “Add.” Keep going until you’ve recoded all 4 scores, then click “Continue.” This will take you back to the window in Figure 1, and you can simply click “OK.” If you now scroll to the end of your data file, you’ll find five new columns there, one for each of the recoded variables.

Conducting a Reliability Analysis (Cronbach’s α) Now you’re ready to check the reliability of your scale. Go up to the “Analyze” menu and choose “Scale,” then “Reliability Analysis…” You should see a window like the one below:

Figure 3

Psychology 280: Psychological Statistics Fall 2005 Borton & Pierce p. 3

Place the variables you want into the box labeled “Items” by clicking on each one in the list on the left and then clicking on the arrow. Make sure to include the recoded items instead of the originals, where appropriate. When all variables have been entered, click the box marked “Statistics.” You’ll get a window like the one below:

Figure 4 In the upper right-hand corner, check the box labeled Inter-Item Correlations. This will give you a correlation matrix for all items in your scale, allowing you to look at how highly each item correlates with each other item. In the upper left-hand corner, click in all 3 boxes under “Descriptives for” (i.e., scale, item, and scale if item deleted). This will give you the mean and standard deviation for each item, for the scale as a whole, and for the whole scale if each item were to be deleted. In addition, you will get Cronbach’s alpha, which I described above. A high alpha (.7 and higher) would be consistent with your hypothesis that all of your scale items are measuring the same construct. If you have a low alpha (below .7), check the column of your output labeled “Cronbach’s alpha if item deleted.” Would your alpha go up if you got rid of any items? Which item(s)? You should check the column labeled “Corrected item-total correlations.” Note that items with low item-total correlations are the ones that would allow you to increase the scale’s alpha by deleting them. Go to your questionnaire to take a look at the items with low item-total correlations. Can you see what might be wrong with these items or why they might not be measuring what all the other items are measuring? Use your judgment in determining which items to delete. If deleting an item increases the alpha only by a miniscule amount, don’t delete it. Using the Compute Command to Create a Mean Score on Your Scale (Creating a Composite Variable) Once you have determined which items should remain in your final scale, you will need to compute what is known as a “composite variable” for the scale. That is, you will compute a variable that represents each participant’s mean of all the scale items. Let’s return to our example of the Rosenberg Self-Esteem Scale. The scale has 10 items, so if I wanted to compute each participant’s mean self-esteem score based on those 10 items, I could use a compute statement. Don’t forget that we need to use the recoded items instead of the originals for questions 3, 5, 8, 9, and 10. To create our compute statement, we would go up to the Transform menu again, and scroll down to “Compute.” You should then see a window that looks like the one below:

Psychology 280: Psychological Statistics Fall 2005 Borton & Pierce p. 4

Figure 3

Note how I’ve typed “meanse” in the “Target Variable” box. That means I want SPSS to create a new column labeled “meanse” that will represent each participant’s mean on the self-esteem scale. In the “Numeric Expression” box, I’ve indicated that I want this new variable to equal the mean of the items on the self-esteem scale, substituting the recoded items for the originals where appropriate. Note that you can use the compute command to perform any of a number of functions. I’ve chosen to calculate a mean. [Note: If you wanted to save this formula to use again later, you could click “Paste,” and the formula would get pasted into the Syntax window, where you could run the command but also save it.] When you’re all set, click “OK.” If we scrolled to the end of the data file containing the SE scores, we’d see a new column labeled “meanse.” This variable would represent each participant’s average SE score based on the 10-item scale. NOTE: In class, each group will calculate Cronbach’s α for its own scale, and will decide which scale items to retain in the final scale. Because all groups will need to have composite variables for each of the four scales, we will ask each group to write on the board which items were included in the final scale and what the final Cronbach’s α was. That way, everyone will have all four alphas to report in their papers, and each group can then create a composite variable for each of the four scales using the Compute command. Part II: Assessing the Convergent and Discriminant Validity of Your Questionnaire Before you begin these analyses, you should already have decided, based on your nomological net, which scale(s) should correlate most and least highly with your own. Once you have made these hypotheses, you are ready to begin examining the correlations among the scales. Remember to first check out a scatterplot before calculating Pearson’s r. To do so, you go under the “Graphs” menu and choose “Scatter/Dot...” as you did for HW 4. If the relationship appears to be linear, you are ready to examine the correlations. Recall that to do this, you go to the “Analyze” menu, scroll down to “Correlate,” and choose “Bivariate.” When looking at the correlation matrix in your output window, remember that the top number represents Pearson’s r (the correlation), the “Sig (2-tailed)” value is the p value (which will be < .05 if the two variables are statistically significantly correlated with one another), and the bottom number is N (sample size).